MODEM: A Morton-Order Degradation Estimation Mechanism for Adverse Weather Image Recovery
Hainuo Wang, Qiming Hu, Xiaojie Guo · 2025
Read MODEM: A Morton-Order Degradation Estimation Mechanism for Adverse Weather Image Recovery by Hainuo Wang, Qiming Hu, Xiaojie Guo (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-17581.txt.
Introduction
Hainuo Wang and co-authors restore photos spoiled by bad weather, in work from 2025. Rain streaks mar small zones, while haze veils whole scenes, so one fix rarely fits all. Their system, MODEM, first gauges the damage, then repairs with that gauge as a guide.
A scan module reads the image in a space-filling order that keeps near and far links in view. A twin gauge splits harm into global and local parts, and those gauges tune the repair steps live. Tests across many weather types and data sets set new best scores. The authors plan to release their code in public.
Catalog record
| CREATOR | Hainuo Wang, Qiming Hu, Xiaojie Guo |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 69 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-17581
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
MODEM: A Morton-Order Degradation Estimation Mechanism for Adverse Weather Image Recovery Hainuo Wang, Qiming Hu, Xiaojie Guo Abstract Restoring images degraded by adverse weather remains a significant challenge due to the highly non-uniform and spatially heterogeneous nature of weather-induced artifacts, e.g. , fine-grained rain streaks versus widespread haze. Accurately estimating the underlying degradation can intuitively provide restoration models with more targeted and effective guidance, enabling adaptive processing strategies. To this end, we propose a Morton-Order Degradation Estimation Mechanism (MODEM) for adverse weather image restoration. Central to MODEM is the Morton-Order 2D-Selective-Scan Module (MOS2D), which integrates Morton-coded spatial ordering with selective state-space models to capture long-range dependencies while preserving local structural coherence. Complementing MOS2D, we introduce a Dual Degradation Estimation Module (DDEM) that disentangles and estimates both global and local degradation priors. These priors dynamically condition the MOS2D modules, facilitating adaptive and context-aware restoration. Extensive experiments and ablation studies demonstrate that MODEM achieves state-of-the-art results across multiple benchmarks and weather types, highlighting its effectiveness in modeling complex degradation dynamics. Our code will be released at here . 1 Introduction Computer vision systems are increasingly integral to critical applications, such as autonomous driving [ 51 , 1 ] and intelligent surveillance [ 49 , 70 ] , demanding reliable performance in diverse environments. However, their effectiveness deteriorates significantly under adverse weather conditions, such as rain [ 16 , 42 , 57 , 75 , 78 , 92 ] , haze [ 76 , 77 , 21 , 62 , 71 , 83 , 85 ] , and snow [ 38 , 45 , 59 , 87 , 13 , 12 ] , which introduce complex visual artifacts and obscure critical scene information. Thus, effectively restoring clear images from such weather-degraded inputs is essential to boost the robustness and safety of modern computer vision technologies in real-world deployments. Early task-specific methods [ 2 , 4 , 23 , 60 , 61 , 30 , 32 , 90 ] largely rely on physical models or hand-crafted statistical priors tailored to individual weather phenomena. Due to the limited feature representation capabilities, these schemes are brittle in the face of complex scenes or deviations from assumed degradation patterns. With the advent of deep learning, numerous deep networks have achieved impressive performance by training on large-scale datasets for specific tasks such as image deraining [ 16 , 42 , 57 , 75 , 78 , 5 , 41 , 69 , 83 ] , dehazing [ 76 , 77 , 21 , 62 , 71 , 83 , 85 ] , or desnowing [ 38 , 45 , 59 , 87 ] . These models excel at implicitly learning the inverse mapping through extensive supervision. But their highly specialized nature necessitates separate models for each weather condition, severely limiting scalability and practical deployment in unconstrained environments. (a) Raster Scanning (b) Continuous Scanning (c) Local Scanning (d) Morton Scanning Figure 1 : Comparison of various image scanning methods. (a) Raster Scanning. (b) Continuous Scanning. (c) Local Scanning. (d) Morton Scanning, which can better preserve spatial locality of neighboring pixels in the resulting 1D sequence, beneficial for capturing contextual information. Recently, much attention has been directed toward unified or multi-task frameworks designed to cope with various weather conditions within a single model [ 95 , 67 , 11 , 40 , 55 , 65 , 79 ] . These approaches adopt a variety of sophisticated strategies, including architecture search for optimal generalization [ 40 ] , disentanglement of weather-shared and weather-specific representations [ 95 ] , knowledge distillation and regularization [ 11 ] , generative diffusion modeling [ 55 ] , learned priors or codebooks [ 79 ] , and transformer-based architectures that offer strong global modeling capabilities [ 67 , 65 ] . Although these models mark notable progress and offer broader applicability, they still struggle with the challenge of modeling the inherently distinct and often highly spatially heterogeneous characteristics of different weather degradations. For example, haze tends to cause smooth intensity attenuation, whereas rain streaks and snowflakes introduce sharp, local occlusions with specific structural patterns. Most existing architectures lack dedicated degradation estimation mechanisms to explicitly model and leverage these fine-grained, spatially variant degradation patterns, which undermines their performance in real-world deployment. Therefore, effective estimation of spatially variant degradation characteristics, encompassing both global trends and local structures, as guidance is critical for adaptive and context-aware restoration under dynamic weather conditions . For an element ( e.g. , pixel) of a degraded image, its degradation characteristic can be regarded as a latent “state" within a degradation space, encapsulating the influence of adverse weather on its visual appearance. These states evolve spatially, governed by both local ( e.g. , rain streaks) and non-local patterns ( e.g. , drifting fog). This perspective naturally lends itself to State Space Modeling (SSM) [ 17 ] . The connection between the image degradation modeling and SSM is formally given in Sec. 3 . By treating degradation as a structured sequence of evolving states, SSM can capture long-range contextual dependencies while maintaining sensitivity to local features. When coupled with degradation estimation mechanisms, SSM empowers the restoration process to be contextually aware and spatially adaptive, effectively aligning restoration strategies with the heterogeneous and scene-specific nature of weather-induced artifacts. Motivated by the above insight, we propose a novel Morton-Order Degradation Estimation Mechanism (MODEM) that estimates degradation state via SSM and guides adaptive restoration in degraded images. At the heart of MODEM lies the Morton-Order 2D-Selective-Scan Module (MOS2D), which integrates SSM with Morton-coded spatial traversal. Unlike conventional raster scanning, the Morton-order scan follows a hierarchical, locality-preserving sequence [ 50 , 34 , 53 , 31 , 7 , 54 ] that enables structured and efficient modeling of both local and long-range dependencies, as shown in Fig. 1 (d). To complement MOS2D, we further introduce a Dual Degradation Estimation Module (DDEM), designed to extract two complementary forms of degradation information from the input: (i) a global degradation representation that encapsulates high-level weather characteristics such as type and severity, and (ii) a set of spatially adaptive kernels that encode local degradation structures and variations. These dual degradation representations are then utilized to dynamically modulate the restoration process. Specifically, the global representation adaptively influences feature transformations within MODEM, while the spatially adaptive kernels guide the matrix construction within MOS2D, refining the spatial dependencies captured along the Morton-order sequence. This dual-modulation strategy equips the restoration network with both global awareness and local sensitivity, thereby delivering more precise and effective restoration. Our primary contributions can be summarized as follows: • We propose a novel Morton-Order Degradation Estimation Mechanism that introduces the MOS2D module, which integrates Morton-coded spatial ordering with selective state space modeling to effectively capture spatially heterogeneous weather degradation dynamics. • We design a Dual Degradation Estimation Module that jointly estimates global degradation descriptors and spatially adaptive kernels. These two complementary representations are used to dynamically modulate MOS2D, allowing contextually aware and spatially adaptive restoration. • Extensive experiments and ablation studies are conducted to demonstrate the effectiveness of the proposed MODEM, and its superiority over other state-of-the-art competitors in restoring images under diverse and complex adverse weather conditions. 2 Related Work This section briefly reviews representative approaches to single adverse weather restoration